Variable Selection Method for Aluminum Electrolytic Process Based on FNN and RM in KPLS Feature Space

Selecting the process variables is an important prerequisite for establishing an accurate model of aluminum electrolytic process. A variable selection method is researched and proposed based on the False Nearest Neighbors (FNN) and Randomization Method (RM) (FR) in KPLS(Kernel Partial Least Squares) feature space. Firstly, the KPLS is employed to transform the original space to the PLS feature space; secondly, in the new feature space, the FNN is used to calculate the similarity measure of each variable which first is retained and then reset to zero for evaluateing the importance to the dependent variable; then, the RM is utilized to test the significance level of the importance for each variable in turn, so the redundant variables would be excluded; lastly, technical energy consumption model of aluminum electrolytic process is built to verify the presented method. The experimental results show that the method selects out the best process variables of aluminum electrolytic process. Therefore, the research provides a new method of the variable selection for metallurgical industrial processes.